The policy is a selective adaptation layered onto a substantially imported architecture.
South Africa's AI policy trajectory represents a hybrid effort, but one whose centre of gravity sits closer to the derivative end of the spectrum than to genuine origination. The 2026 Draft Policy makes real and sometimes inventive attempts at contextual grounding: anchoring AI governance in the Constitution, invoking Ubuntu, and foregrounding indigenous language preservation.
However, the core policy architecture (risk-based classification, ethical principles, and innovation-competitiveness framing) remains fundamentally imported.
The result is a document that reads global and annotates local, rather than one that thinks from local conditions outward.
While the current SA
Draft Policy relies heavily on imported, top-heavy institutional blueprints, a genuinely contextual AI
strategy requires ground-level execution.
The following seven pillars outline my recommended
framework for how South Africa must translate global ideals into actionable policies across
National, Provincial, and Local tiers.
AI policy is ultimately a question of whose life it touches. The citizen is not a passive data subject but a current or possible innovator and rights-bearer entitled to explanation, redress, privacy, and meaningful participation under the Constitution.
No AI policy survives contact with an unstable grid, thin fibre coverage, or foreign-owned compute. Infrastructure is a hard constraint.
A risk-classification regime borrowed from Brussels presumes a regulator that can audit, inspect, and enforce. A deep institutional capacity check is required.
Over a third of South Africa's workforce operates outside the formal economy. An AI policy that speaks only to large firms writes millions out.
Twelve official languages and indigenous knowledge systems determine whether AI is built for South Africans or merely deployed on them.
Without a clear inter-governmental architecture, the framework risks becoming a national document that never reaches the municipal coalface.
Innovation is treated as an outcome if ethics are addressed, missing its role in political-economic design and local knowledge structures.
The Framework nowhere commits the state to building a sovereign foundation model or to reorganising the machinery of government around AI-enabled decision-making. It names every argument for doing so (data sovereignty, foreign-provider dependence, indigenous-language under-representation), then declines to prescribe the treatment. The document imagines AI as something the state will oversee, not something the state will become. This is the difference between being a consumer and a producer of AI, and the Framework, for all its ambition, sits squarely on the consumer side.
Key structural similarities and differences across the international texts and the SA frameworks.
Convergence of Terminology: Terms such as "human-centric," "trustworthy AI," "transparency," "accountability," "fairness," and "risk-based approach" appear across every text. This reflects a unified language template originating primarily from the OECD AI Principles and UNESCO.
Risk-based classification forms the structural backbone for the EU AI Act, the UN Model Policy, and the SA Draft Policy. Each relies on regulating through harm grading.
Governance architectures display mirrored elements: dedicated oversight bodies, ethics boards, and audit obligations are pervasive globally.
The EU AI Act is sweeping binding legislation. The SA Draft Policy is a framework that may lead to legislation down the line. The UN documents remain purely guidance mechanisms.
Only the SA documents repeatedly frame AI primarily as a development tool destined to address poverty, unemployment, and inequality.
The SA documents explicitly invoke Ubuntu, setting them conceptually apart from the purely individual rights-based structure of the EU.
Both SA and the AU frameworks share a definitive developmental orientation, emphasizing the "AI divide" and harnessing technical benefits amidst historical constraints.
The AU Strategy is aggressive about African-centric AI capability and cultural renaissance. SA acknowledges these but predominantly leans on OECD regulatory vernacular rather than fully embracing continental disruption.
Evaluating the authoritative levers and core narrative mechanisms driving the policy documents.
Diagnostic breakdown of specific thematic modules within the policy and their degree of local grounding.
| Theme | Classification | Analytic Take |
|---|---|---|
| Risk-Based Regulation | Imported | Overtly extracts the EU AI Act's structural logic globally transposed. |
| Ethical Principles | Imported | Nearly verbatim mimicry of prevailing OECD and major corporate principles. |
| Inclusion & Equity | Adapted | Integrates SA's legacy of inequality into global inclusiveness dialogue. |
| Constitutional Anchoring | Grounded | Specifically intertwined with provisions from the SA Bill of Rights and domestic tech law. |
| AI Insurance Superfund | Grounded | Adapted off the framework of the localized Road Accident Fund concept. |
| Ubuntu Philosophy | Performative | Invoked purely on an aspirational values basis without structural mechanics. |
| Education System | Performative | Ignores baseline disparities in the current school system while projecting AI ubiquity. |
| Digital Infrastructure | Ambiguous | Recommends vast technological networking completely detached from persistent grid instability. |
| Informal Economy | Unsaid | Total omission of an economic engine involving millions of marginalized participants. |
| Fiscal Feasibility | Unsaid | Broad mandates built on absent budgeting metrics. |
A framework defined as much by its exclusions as its written ambitions.
The policy mandates creating seven brand-new institutional regulatory bodies-all while existent regulators face overwhelming operational constraint. Institutional blueprinting frequently overshadows execution capabilities.
Data centres, digital sovereignty, and complex computing architectures run sharply counter to South Africa's deep realities regarding load shedding and electrical grid decay. This critical constraint is profoundly understated.
A staggering percentage of SA's workforce relies on the informal economy. And yet, this sector has no seat at the table concerning automation vectors, economic shockwaves, or transitional protections driven by AI propagation.
Nowhere in the Framework does the state commit to building a national foundation model. It worries about data sovereignty, foreign providers, and the under-representation of South African languages, and then proposes no domestic model to address any of it. The UAE built Falcon, France backed Mistral, Singapore funded SEA-LION, India is funding Indic LLMs; South Africa regulates and enables but does not build. The document names every disease the treatment would cure, and then declines to prescribe it.
The Framework proposes many new bodies (an Ethics Board, a National AI Commission, a Regulatory Authority, a Safety Institute, an Ombudsperson), yet all of them sit above the machinery of government, watching it. None are inside the departments rebuilding how decisions are actually made. Serious AI-driven governance implies central digital services with real authority, data stewards embedded in ministries, rewritten procurement law, and new civil-service training. The document treats AI as a tool to procure rather than an occasion to reconstruct public capacity.
South Africa is actively engaging in policy creation, translating modern global AI governance discourse into its local institutional syntax. This constitutes a valuable but undeniably derivative enterprise.
The profound and particular realities of the nation-informal economics, profound energy deficits, structural limits in fiscal policy, and immense bureaucratic stresses-must generate the shape of policy formulation. Allowing core values like Ubuntu to manifest as concrete operational tools will begin defining a grounded, rather than merely annotated, national path forward.
1. Republic of South Africa, Department of Communications and Digital Technologies. National Artificial Intelligence Policy Framework. Government Gazette No. 54477, 10 April 2026. Pretoria: Government Printing Works.
2. On the economics of scraping and journalism, see Reuters Institute for the Study of Journalism, Digital News Report (annual), and contemporary debate on text-and-data mining exceptions in copyright law.
3. African Union Commission. Continental Artificial Intelligence Strategy: Harnessing AI for Africa's Development and Prosperity. Addis Ababa, July 2024.
4. Pritchett, L., Woolcock, M., & Andrews, M. (2013). Looking Like a State: Techniques of Persistent Failure in State Capability for Implementation. Journal of Development Studies, 49(1), 1–18.
5. Andrews, M., Pritchett, L., & Woolcock, M. (2017). Building State Capability: Evidence, Analysis, Action. Oxford: Oxford University Press.
6. Supplementary: UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021); OECD AI Principles (2019, updated 2024); European Parliament, Regulation (EU) 2024/1689 (AI Act).